■
[ AIR-GAPPED ] 100% ON-PHONE SILICON
AI chat that stays local.
No cloud, no tracking.
Run language models directly on your phone with llama.cpp. No cloud servers, no account required, zero network telemetry.
NETWORK: AIR-GAPPED (0 BYTES EGRESS)
/
RUNTIME: LLAMA.CPP C++20
/
FORMAT: GGUF QUANTIZED
ARCHITECTURE // SYSTEM_PRINCIPLES
VERIFIED
01 // RUNTIME
Local Inference
Executes quantized GGUF weights directly on local silicon via llama.cpp. Zero remote API dependencies.
02 // TELEMETRY
Hardware Benchmarks
Profiles physical execution performance on your phone: time-to-first-token (TTFT), token throughput, and memory consumption.
03 // ISOLATION
Absolute Privacy
No cloud accounts, no analytics SDKs, no tracking cookies. Works completely in airplane mode with Wi-Fi disabled.
FREQUENTLY_ASKED_QUESTIONS
4 ENTRIES
[+] What is Pokke? [EXPAND]
Pokke is a sovereign on-phone AI chat application and mobile hardware benchmark platform developed by Auralis Studio. It allows you to run local language models directly on your phone without transmitting prompt or conversation data to the cloud.
[+] Which models can it run? [EXPAND]
Pokke supports standard GGUF quantized models (such as Llama, Qwen, Gemma, and Phi) optimized for mobile and edge silicon.
[+] Does it require an internet connection? [EXPAND]
No. Pokke is designed to be completely air-gapped. Once model weights are loaded on your phone, all inference, benchmarking, and chat history management occur entirely offline with zero network connectivity.
[+] How does on-phone inference protect privacy? [EXPAND]
Because model execution happens entirely inside your phone's memory using llama.cpp, your prompts, system instructions, and generated tokens never leave your phone. There are no cloud intermediaries, analytics trackers, or remote telemetry sockets.